ENTITY

Alan Turing Institute

Properties1
tagsresearch, data-science, ai, data-poisoning

1 min read

Alan Turing Institute

Type: organization (UK national research institute) | Domain relevance: Co-author of the data-poisoning study behind the wiki’s “250 documents” correction.

Summary

The Alan Turing Institute is the UK’s national institute for data science and AI. It co-conducted the “small samples poison LLMs” study with Anthropic and the UK AI Security Institute. Along with the other co-authors, its participation is why the study’s finding — ~250 documents can backdoor an LLM — is credible research, even though the SEO application (250 Authority Protocol) misuses it (see Data Poisoning & LLM Backdoors).

Key Facts

  • UK national institute for data science and AI.
  • Co-authored the 250-documents poisoning study.
  • Adds academic credibility to the finding.

In the Sources

SourceContext
A small number of samples can poison LLMs of any sizeStudy co-author (with Anthropic + UK AISI).

Entities: Anthropic, UK AI Security Institute Concepts: Data Poisoning & LLM Backdoors, 250 Authority Protocol